Research Of License Plate Recognition System Under Complicated Illumination

License plate recognition as an important part of intelligent transportation system plays an important role in real-time, accurate, and efficient transportation management system and becomes the main direction of the current traffic management research because of its capacity of perceiving evehicle uniqueness.Research of the existing license plate recognition system done in this paper shows that most existing detection methods are still positioned to varying degrees by many of their environment control system, etc after contrasting variety of location detection method, Especially in the face of insufficient light or too strong, the lower the image quality vehicles collected, often directly affect the outcome of the license plate location which is not conducive to the license plate recognition.This paper presents the case of license plate recognition system design under complex illumination which is based on computer vision recognition technology, several key technologies of license plate recognition system have been studied and improved, such as:image enhancement, image segmentation, license plate binarization algorithm, character segmentation and character recognition.License plate recognition system major module and methods described here description:In the image pre-processing module, complexity of light (uneven lighting, light abnormal reflex, etc.) ofter leads to serious degradation of license plate image and the binarization algorithm is often unsatisfactory which affect subsequent processing. In this paper, the improved level cap transformation of the traditional fixed threshold binarization algorithm has been optimized and applied to the license plate image binarization algorithm, improving the impact of non-uniform illumination of license plate images.In the license plate location module, the circumstances of plate image distortion decide which specific positioning method is used,when the image color distortion is not serious, we detect it using license plate location method based on fuzzy adaptive Wiener filtering. For detailed,we take advantage of HSV space to extract plate image saturation level, then adjust the adaptive Wiener filter filtering window so as to obfuscate the image of the Non-plate area, finally we use morphological processing methods to achieve the ultimate license plate positioning. Otherwise, on the basis of obtaining a clearer binary image picture, we use pretreatment method based on morphology to calculate the edges of the image projected area, look the valley point, and roughly determine the location of the license plate, and then calculate the aspect ratio of the connected domain, excluding connected domain field value is not within range, and finally get the license plate area.In character segmentation module, this paper presents a differential projection and excellent cutting character license plate character segmentation the main method of which is to use level difference projection to achieve tilt correction and horizontal cut,and to find excellent cut characters of the license plate to accomplish character vertical split,combined with the connected domain and histogram projection method. The algorithm Effectively solve the character adhesions and the character recognition when characters fracture.In character recognition module, this paper adopts template matching method which is three more commonly used methods with mathematical statistics-based method and artificial intelligence techniques and increase the customer template customization function of Chinese characters in order to make the software application sides are able to increase emerging license plate characters by themselves. Meanwhile the Kanji multi-template function greatly improves the recognition rate of Chinese characters and provides confidence parameter characters of kanji numbers and letters output identification results.Moreover blur detection function and visual intelligence function can be easily added to the software.The system designed in this paper has been realized on the matlab R2009a platform. Experimental verification indicated that the recognition system can effectively solve the impact of complex environmental on the license plate image, and is provided with higher accuracy and faster recognition speed compared to color-based license plate recognition systems and multi-image texture features-based license license recognition system.